11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * fs/fs-writeback.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Contains all the functions related to writing back and waiting 71da177e4SLinus Torvalds * upon dirty inodes against superblocks, and writing back dirty 81da177e4SLinus Torvalds * pages against inodes. ie: data writeback. Writeout of the 91da177e4SLinus Torvalds * inode itself is not handled here. 101da177e4SLinus Torvalds * 11e1f8e874SFrancois Cami * 10Apr2002 Andrew Morton 121da177e4SLinus Torvalds * Split out of fs/inode.c 131da177e4SLinus Torvalds * Additions for address_space-based writeback 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 161da177e4SLinus Torvalds #include <linux/kernel.h> 17f5ff8422SJens Axboe #include <linux/module.h> 181da177e4SLinus Torvalds #include <linux/spinlock.h> 195a0e3ad6STejun Heo #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/sched.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/mm.h> 2303ba3782SJens Axboe #include <linux/kthread.h> 2403ba3782SJens Axboe #include <linux/freezer.h> 251da177e4SLinus Torvalds #include <linux/writeback.h> 261da177e4SLinus Torvalds #include <linux/blkdev.h> 271da177e4SLinus Torvalds #include <linux/backing-dev.h> 281da177e4SLinus Torvalds #include <linux/buffer_head.h> 2907f3f05cSDavid Howells #include "internal.h" 301da177e4SLinus Torvalds 3166f3b8e2SJens Axboe #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) 32f11b00f3SAdrian Bunk 3303ba3782SJens Axboe /* 34d0bceac7SJens Axboe * We don't actually have pdflush, but this one is exported though /proc... 35d0bceac7SJens Axboe */ 36d0bceac7SJens Axboe int nr_pdflush_threads; 37d0bceac7SJens Axboe 38d0bceac7SJens Axboe /* 39c4a77a6cSJens Axboe * Passed into wb_writeback(), essentially a subset of writeback_control 40c4a77a6cSJens Axboe */ 41c4a77a6cSJens Axboe struct wb_writeback_args { 42c4a77a6cSJens Axboe long nr_pages; 43c4a77a6cSJens Axboe struct super_block *sb; 44c4a77a6cSJens Axboe enum writeback_sync_modes sync_mode; 4552957fe1SH Hartley Sweeten unsigned int for_kupdate:1; 4652957fe1SH Hartley Sweeten unsigned int range_cyclic:1; 4752957fe1SH Hartley Sweeten unsigned int for_background:1; 48c4a77a6cSJens Axboe }; 49c4a77a6cSJens Axboe 50c4a77a6cSJens Axboe /* 5103ba3782SJens Axboe * Work items for the bdi_writeback threads 52f11b00f3SAdrian Bunk */ 5303ba3782SJens Axboe struct bdi_work { 548010c3b6SJens Axboe struct list_head list; /* pending work list */ 558010c3b6SJens Axboe struct rcu_head rcu_head; /* for RCU free/clear of work */ 5603ba3782SJens Axboe 578010c3b6SJens Axboe unsigned long seen; /* threads that have seen this work */ 588010c3b6SJens Axboe atomic_t pending; /* number of threads still to do work */ 5903ba3782SJens Axboe 608010c3b6SJens Axboe struct wb_writeback_args args; /* writeback arguments */ 6103ba3782SJens Axboe 628010c3b6SJens Axboe unsigned long state; /* flag bits, see WS_* */ 6303ba3782SJens Axboe }; 6403ba3782SJens Axboe 6503ba3782SJens Axboe enum { 667f0e7bedSChristoph Hellwig WS_INPROGRESS = 0, 677f0e7bedSChristoph Hellwig WS_ONSTACK, 6803ba3782SJens Axboe }; 6903ba3782SJens Axboe 7003ba3782SJens Axboe static inline void bdi_work_init(struct bdi_work *work, 71b6e51316SJens Axboe struct wb_writeback_args *args) 7203ba3782SJens Axboe { 7303ba3782SJens Axboe INIT_RCU_HEAD(&work->rcu_head); 74b6e51316SJens Axboe work->args = *args; 757f0e7bedSChristoph Hellwig __set_bit(WS_INPROGRESS, &work->state); 7603ba3782SJens Axboe } 7703ba3782SJens Axboe 78f11b00f3SAdrian Bunk /** 79f11b00f3SAdrian Bunk * writeback_in_progress - determine whether there is writeback in progress 80f11b00f3SAdrian Bunk * @bdi: the device's backing_dev_info structure. 81f11b00f3SAdrian Bunk * 8203ba3782SJens Axboe * Determine whether there is writeback waiting to be handled against a 8303ba3782SJens Axboe * backing device. 84f11b00f3SAdrian Bunk */ 85f11b00f3SAdrian Bunk int writeback_in_progress(struct backing_dev_info *bdi) 86f11b00f3SAdrian Bunk { 8703ba3782SJens Axboe return !list_empty(&bdi->work_list); 88f11b00f3SAdrian Bunk } 89f11b00f3SAdrian Bunk 9003ba3782SJens Axboe static void bdi_work_free(struct rcu_head *head) 914195f73dSNick Piggin { 9203ba3782SJens Axboe struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); 934195f73dSNick Piggin 947f0e7bedSChristoph Hellwig clear_bit(WS_INPROGRESS, &work->state); 957f0e7bedSChristoph Hellwig smp_mb__after_clear_bit(); 967f0e7bedSChristoph Hellwig wake_up_bit(&work->state, WS_INPROGRESS); 977f0e7bedSChristoph Hellwig 987f0e7bedSChristoph Hellwig if (!test_bit(WS_ONSTACK, &work->state)) 9903ba3782SJens Axboe kfree(work); 1001da177e4SLinus Torvalds } 1011da177e4SLinus Torvalds 10203ba3782SJens Axboe static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) 10303ba3782SJens Axboe { 1041da177e4SLinus Torvalds /* 10503ba3782SJens Axboe * The caller has retrieved the work arguments from this work, 10603ba3782SJens Axboe * drop our reference. If this is the last ref, delete and free it 10703ba3782SJens Axboe */ 10803ba3782SJens Axboe if (atomic_dec_and_test(&work->pending)) { 10903ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 11003ba3782SJens Axboe 11103ba3782SJens Axboe spin_lock(&bdi->wb_lock); 11203ba3782SJens Axboe list_del_rcu(&work->list); 11303ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 11403ba3782SJens Axboe 1157f0e7bedSChristoph Hellwig call_rcu(&work->rcu_head, bdi_work_free); 11603ba3782SJens Axboe } 11703ba3782SJens Axboe } 11803ba3782SJens Axboe 11903ba3782SJens Axboe static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) 12003ba3782SJens Axboe { 12103ba3782SJens Axboe work->seen = bdi->wb_mask; 12203ba3782SJens Axboe BUG_ON(!work->seen); 12303ba3782SJens Axboe atomic_set(&work->pending, bdi->wb_cnt); 12403ba3782SJens Axboe BUG_ON(!bdi->wb_cnt); 12503ba3782SJens Axboe 12603ba3782SJens Axboe /* 127deed62edSNick Piggin * list_add_tail_rcu() contains the necessary barriers to 128deed62edSNick Piggin * make sure the above stores are seen before the item is 129deed62edSNick Piggin * noticed on the list 1301da177e4SLinus Torvalds */ 13103ba3782SJens Axboe spin_lock(&bdi->wb_lock); 13203ba3782SJens Axboe list_add_tail_rcu(&work->list, &bdi->work_list); 13303ba3782SJens Axboe spin_unlock(&bdi->wb_lock); 1341da177e4SLinus Torvalds 1351da177e4SLinus Torvalds /* 13603ba3782SJens Axboe * If the default thread isn't there, make sure we add it. When 13703ba3782SJens Axboe * it gets created and wakes up, we'll run this work. 1381da177e4SLinus Torvalds */ 13903ba3782SJens Axboe if (unlikely(list_empty_careful(&bdi->wb_list))) 14003ba3782SJens Axboe wake_up_process(default_backing_dev_info.wb.task); 14103ba3782SJens Axboe else { 14203ba3782SJens Axboe struct bdi_writeback *wb = &bdi->wb; 1431da177e4SLinus Torvalds 1441ef7d9aaSNick Piggin if (wb->task) 14503ba3782SJens Axboe wake_up_process(wb->task); 1461da177e4SLinus Torvalds } 14703ba3782SJens Axboe } 1481da177e4SLinus Torvalds 1491da177e4SLinus Torvalds /* 15003ba3782SJens Axboe * Used for on-stack allocated work items. The caller needs to wait until 15103ba3782SJens Axboe * the wb threads have acked the work before it's safe to continue. 1521da177e4SLinus Torvalds */ 1537f0e7bedSChristoph Hellwig static void bdi_wait_on_work_done(struct bdi_work *work) 1541da177e4SLinus Torvalds { 1557f0e7bedSChristoph Hellwig wait_on_bit(&work->state, WS_INPROGRESS, bdi_sched_wait, 15603ba3782SJens Axboe TASK_UNINTERRUPTIBLE); 15703ba3782SJens Axboe } 15803ba3782SJens Axboe 159f11fcae8SJens Axboe static void bdi_alloc_queue_work(struct backing_dev_info *bdi, 160f17625b3SJens Axboe struct wb_writeback_args *args) 16103ba3782SJens Axboe { 16203ba3782SJens Axboe struct bdi_work *work; 16303ba3782SJens Axboe 164bcddc3f0SJens Axboe /* 165bcddc3f0SJens Axboe * This is WB_SYNC_NONE writeback, so if allocation fails just 166bcddc3f0SJens Axboe * wakeup the thread for old dirty data writeback 167bcddc3f0SJens Axboe */ 16803ba3782SJens Axboe work = kmalloc(sizeof(*work), GFP_ATOMIC); 169bcddc3f0SJens Axboe if (work) { 170b6e51316SJens Axboe bdi_work_init(work, args); 171f11fcae8SJens Axboe bdi_queue_work(bdi, work); 172bcddc3f0SJens Axboe } else { 173bcddc3f0SJens Axboe struct bdi_writeback *wb = &bdi->wb; 174bcddc3f0SJens Axboe 175bcddc3f0SJens Axboe if (wb->task) 176bcddc3f0SJens Axboe wake_up_process(wb->task); 177bcddc3f0SJens Axboe } 17803ba3782SJens Axboe } 17903ba3782SJens Axboe 180b6e51316SJens Axboe /** 1813c4d7165SChristoph Hellwig * bdi_queue_work_onstack - start and wait for writeback 182b6e51316SJens Axboe * @sb: write inodes from this super_block 183b6e51316SJens Axboe * 184b6e51316SJens Axboe * Description: 1853c4d7165SChristoph Hellwig * This function initiates writeback and waits for the operation to 1863c4d7165SChristoph Hellwig * complete. Callers must hold the sb s_umount semaphore for 187b6e51316SJens Axboe * reading, to avoid having the super disappear before we are done. 18803ba3782SJens Axboe */ 1893c4d7165SChristoph Hellwig static void bdi_queue_work_onstack(struct wb_writeback_args *args) 190b6e51316SJens Axboe { 191f0fad8a5SChristoph Hellwig struct bdi_work work; 192f0fad8a5SChristoph Hellwig 1933c4d7165SChristoph Hellwig bdi_work_init(&work, args); 1947f0e7bedSChristoph Hellwig __set_bit(WS_ONSTACK, &work.state); 195f0fad8a5SChristoph Hellwig 1963c4d7165SChristoph Hellwig bdi_queue_work(args->sb->s_bdi, &work); 1977f0e7bedSChristoph Hellwig bdi_wait_on_work_done(&work); 19803ba3782SJens Axboe } 199b6e51316SJens Axboe 200b6e51316SJens Axboe /** 201b6e51316SJens Axboe * bdi_start_writeback - start writeback 202b6e51316SJens Axboe * @bdi: the backing device to write from 2034b6764faSJaswinder Singh Rajput * @sb: write inodes from this super_block 204b6e51316SJens Axboe * @nr_pages: the number of pages to write 205b6e51316SJens Axboe * 206b6e51316SJens Axboe * Description: 207b6e51316SJens Axboe * This does WB_SYNC_NONE opportunistic writeback. The IO is only 208b6e51316SJens Axboe * started when this function returns, we make no guarentees on 2090e3c9a22SJens Axboe * completion. Caller need not hold sb s_umount semaphore. 210b6e51316SJens Axboe * 211b6e51316SJens Axboe */ 212a72bfd4dSJens Axboe void bdi_start_writeback(struct backing_dev_info *bdi, struct super_block *sb, 2130e3c9a22SJens Axboe long nr_pages) 214b6e51316SJens Axboe { 215b6e51316SJens Axboe struct wb_writeback_args args = { 216a72bfd4dSJens Axboe .sb = sb, 217b6e51316SJens Axboe .sync_mode = WB_SYNC_NONE, 218b6e51316SJens Axboe .nr_pages = nr_pages, 219b6e51316SJens Axboe .range_cyclic = 1, 220b6e51316SJens Axboe }; 221b6e51316SJens Axboe 222d3ddec76SWu Fengguang /* 223d3ddec76SWu Fengguang * We treat @nr_pages=0 as the special case to do background writeback, 224d3ddec76SWu Fengguang * ie. to sync pages until the background dirty threshold is reached. 225d3ddec76SWu Fengguang */ 226d3ddec76SWu Fengguang if (!nr_pages) { 227d3ddec76SWu Fengguang args.nr_pages = LONG_MAX; 228d3ddec76SWu Fengguang args.for_background = 1; 229d3ddec76SWu Fengguang } 230d3ddec76SWu Fengguang 231f17625b3SJens Axboe bdi_alloc_queue_work(bdi, &args); 2321da177e4SLinus Torvalds } 2331da177e4SLinus Torvalds 2341da177e4SLinus Torvalds /* 2356610a0bcSAndrew Morton * Redirty an inode: set its when-it-was dirtied timestamp and move it to the 2366610a0bcSAndrew Morton * furthest end of its superblock's dirty-inode list. 2376610a0bcSAndrew Morton * 2386610a0bcSAndrew Morton * Before stamping the inode's ->dirtied_when, we check to see whether it is 23966f3b8e2SJens Axboe * already the most-recently-dirtied inode on the b_dirty list. If that is 2406610a0bcSAndrew Morton * the case then the inode must have been redirtied while it was being written 2416610a0bcSAndrew Morton * out and we don't reset its dirtied_when. 2426610a0bcSAndrew Morton */ 2436610a0bcSAndrew Morton static void redirty_tail(struct inode *inode) 2446610a0bcSAndrew Morton { 24503ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 2466610a0bcSAndrew Morton 24703ba3782SJens Axboe if (!list_empty(&wb->b_dirty)) { 24866f3b8e2SJens Axboe struct inode *tail; 2496610a0bcSAndrew Morton 25003ba3782SJens Axboe tail = list_entry(wb->b_dirty.next, struct inode, i_list); 25166f3b8e2SJens Axboe if (time_before(inode->dirtied_when, tail->dirtied_when)) 2526610a0bcSAndrew Morton inode->dirtied_when = jiffies; 2536610a0bcSAndrew Morton } 25403ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 2556610a0bcSAndrew Morton } 2566610a0bcSAndrew Morton 2576610a0bcSAndrew Morton /* 25866f3b8e2SJens Axboe * requeue inode for re-scanning after bdi->b_io list is exhausted. 259c986d1e2SAndrew Morton */ 2600e0f4fc2SKen Chen static void requeue_io(struct inode *inode) 261c986d1e2SAndrew Morton { 26203ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 26303ba3782SJens Axboe 26403ba3782SJens Axboe list_move(&inode->i_list, &wb->b_more_io); 265c986d1e2SAndrew Morton } 266c986d1e2SAndrew Morton 2671c0eeaf5SJoern Engel static void inode_sync_complete(struct inode *inode) 2681c0eeaf5SJoern Engel { 2691c0eeaf5SJoern Engel /* 2701c0eeaf5SJoern Engel * Prevent speculative execution through spin_unlock(&inode_lock); 2711c0eeaf5SJoern Engel */ 2721c0eeaf5SJoern Engel smp_mb(); 2731c0eeaf5SJoern Engel wake_up_bit(&inode->i_state, __I_SYNC); 2741c0eeaf5SJoern Engel } 2751c0eeaf5SJoern Engel 276d2caa3c5SJeff Layton static bool inode_dirtied_after(struct inode *inode, unsigned long t) 277d2caa3c5SJeff Layton { 278d2caa3c5SJeff Layton bool ret = time_after(inode->dirtied_when, t); 279d2caa3c5SJeff Layton #ifndef CONFIG_64BIT 280d2caa3c5SJeff Layton /* 281d2caa3c5SJeff Layton * For inodes being constantly redirtied, dirtied_when can get stuck. 282d2caa3c5SJeff Layton * It _appears_ to be in the future, but is actually in distant past. 283d2caa3c5SJeff Layton * This test is necessary to prevent such wrapped-around relative times 2845b0830cbSJens Axboe * from permanently stopping the whole bdi writeback. 285d2caa3c5SJeff Layton */ 286d2caa3c5SJeff Layton ret = ret && time_before_eq(inode->dirtied_when, jiffies); 287d2caa3c5SJeff Layton #endif 288d2caa3c5SJeff Layton return ret; 289d2caa3c5SJeff Layton } 290d2caa3c5SJeff Layton 291c986d1e2SAndrew Morton /* 2922c136579SFengguang Wu * Move expired dirty inodes from @delaying_queue to @dispatch_queue. 2932c136579SFengguang Wu */ 2942c136579SFengguang Wu static void move_expired_inodes(struct list_head *delaying_queue, 2952c136579SFengguang Wu struct list_head *dispatch_queue, 2962c136579SFengguang Wu unsigned long *older_than_this) 2972c136579SFengguang Wu { 2985c03449dSShaohua Li LIST_HEAD(tmp); 2995c03449dSShaohua Li struct list_head *pos, *node; 300cf137307SJens Axboe struct super_block *sb = NULL; 3015c03449dSShaohua Li struct inode *inode; 302cf137307SJens Axboe int do_sb_sort = 0; 3035c03449dSShaohua Li 3042c136579SFengguang Wu while (!list_empty(delaying_queue)) { 3055c03449dSShaohua Li inode = list_entry(delaying_queue->prev, struct inode, i_list); 3062c136579SFengguang Wu if (older_than_this && 307d2caa3c5SJeff Layton inode_dirtied_after(inode, *older_than_this)) 3082c136579SFengguang Wu break; 309cf137307SJens Axboe if (sb && sb != inode->i_sb) 310cf137307SJens Axboe do_sb_sort = 1; 311cf137307SJens Axboe sb = inode->i_sb; 3125c03449dSShaohua Li list_move(&inode->i_list, &tmp); 3135c03449dSShaohua Li } 3145c03449dSShaohua Li 315cf137307SJens Axboe /* just one sb in list, splice to dispatch_queue and we're done */ 316cf137307SJens Axboe if (!do_sb_sort) { 317cf137307SJens Axboe list_splice(&tmp, dispatch_queue); 318cf137307SJens Axboe return; 319cf137307SJens Axboe } 320cf137307SJens Axboe 3215c03449dSShaohua Li /* Move inodes from one superblock together */ 3225c03449dSShaohua Li while (!list_empty(&tmp)) { 3235c03449dSShaohua Li inode = list_entry(tmp.prev, struct inode, i_list); 3245c03449dSShaohua Li sb = inode->i_sb; 3255c03449dSShaohua Li list_for_each_prev_safe(pos, node, &tmp) { 3265c03449dSShaohua Li inode = list_entry(pos, struct inode, i_list); 3275c03449dSShaohua Li if (inode->i_sb == sb) 3282c136579SFengguang Wu list_move(&inode->i_list, dispatch_queue); 3292c136579SFengguang Wu } 3302c136579SFengguang Wu } 3315c03449dSShaohua Li } 3322c136579SFengguang Wu 3332c136579SFengguang Wu /* 3342c136579SFengguang Wu * Queue all expired dirty inodes for io, eldest first. 3352c136579SFengguang Wu */ 33603ba3782SJens Axboe static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) 3372c136579SFengguang Wu { 33803ba3782SJens Axboe list_splice_init(&wb->b_more_io, wb->b_io.prev); 33903ba3782SJens Axboe move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); 34066f3b8e2SJens Axboe } 34166f3b8e2SJens Axboe 342a9185b41SChristoph Hellwig static int write_inode(struct inode *inode, struct writeback_control *wbc) 34366f3b8e2SJens Axboe { 34403ba3782SJens Axboe if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) 345a9185b41SChristoph Hellwig return inode->i_sb->s_op->write_inode(inode, wbc); 34603ba3782SJens Axboe return 0; 34766f3b8e2SJens Axboe } 34808d8e974SFengguang Wu 3492c136579SFengguang Wu /* 35001c03194SChristoph Hellwig * Wait for writeback on an inode to complete. 35101c03194SChristoph Hellwig */ 35201c03194SChristoph Hellwig static void inode_wait_for_writeback(struct inode *inode) 35301c03194SChristoph Hellwig { 35401c03194SChristoph Hellwig DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); 35501c03194SChristoph Hellwig wait_queue_head_t *wqh; 35601c03194SChristoph Hellwig 35701c03194SChristoph Hellwig wqh = bit_waitqueue(&inode->i_state, __I_SYNC); 35858a9d3d8SRichard Kennedy while (inode->i_state & I_SYNC) { 35901c03194SChristoph Hellwig spin_unlock(&inode_lock); 36001c03194SChristoph Hellwig __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); 36101c03194SChristoph Hellwig spin_lock(&inode_lock); 36258a9d3d8SRichard Kennedy } 36301c03194SChristoph Hellwig } 36401c03194SChristoph Hellwig 36501c03194SChristoph Hellwig /* 36601c03194SChristoph Hellwig * Write out an inode's dirty pages. Called under inode_lock. Either the 36701c03194SChristoph Hellwig * caller has ref on the inode (either via __iget or via syscall against an fd) 36801c03194SChristoph Hellwig * or the inode has I_WILL_FREE set (via generic_forget_inode) 36901c03194SChristoph Hellwig * 3701da177e4SLinus Torvalds * If `wait' is set, wait on the writeout. 3711da177e4SLinus Torvalds * 3721da177e4SLinus Torvalds * The whole writeout design is quite complex and fragile. We want to avoid 3731da177e4SLinus Torvalds * starvation of particular inodes when others are being redirtied, prevent 3741da177e4SLinus Torvalds * livelocks, etc. 3751da177e4SLinus Torvalds * 3761da177e4SLinus Torvalds * Called under inode_lock. 3771da177e4SLinus Torvalds */ 3781da177e4SLinus Torvalds static int 37901c03194SChristoph Hellwig writeback_single_inode(struct inode *inode, struct writeback_control *wbc) 3801da177e4SLinus Torvalds { 3811da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 38201c03194SChristoph Hellwig unsigned dirty; 3831da177e4SLinus Torvalds int ret; 3841da177e4SLinus Torvalds 38501c03194SChristoph Hellwig if (!atomic_read(&inode->i_count)) 38601c03194SChristoph Hellwig WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); 38701c03194SChristoph Hellwig else 38801c03194SChristoph Hellwig WARN_ON(inode->i_state & I_WILL_FREE); 38901c03194SChristoph Hellwig 39001c03194SChristoph Hellwig if (inode->i_state & I_SYNC) { 39101c03194SChristoph Hellwig /* 39201c03194SChristoph Hellwig * If this inode is locked for writeback and we are not doing 39366f3b8e2SJens Axboe * writeback-for-data-integrity, move it to b_more_io so that 39401c03194SChristoph Hellwig * writeback can proceed with the other inodes on s_io. 39501c03194SChristoph Hellwig * 39601c03194SChristoph Hellwig * We'll have another go at writing back this inode when we 39766f3b8e2SJens Axboe * completed a full scan of b_io. 39801c03194SChristoph Hellwig */ 399a9185b41SChristoph Hellwig if (wbc->sync_mode != WB_SYNC_ALL) { 40001c03194SChristoph Hellwig requeue_io(inode); 40101c03194SChristoph Hellwig return 0; 40201c03194SChristoph Hellwig } 40301c03194SChristoph Hellwig 40401c03194SChristoph Hellwig /* 40501c03194SChristoph Hellwig * It's a data-integrity sync. We must wait. 40601c03194SChristoph Hellwig */ 40701c03194SChristoph Hellwig inode_wait_for_writeback(inode); 40801c03194SChristoph Hellwig } 40901c03194SChristoph Hellwig 4101c0eeaf5SJoern Engel BUG_ON(inode->i_state & I_SYNC); 4111da177e4SLinus Torvalds 4125547e8aaSDmitry Monakhov /* Set I_SYNC, reset I_DIRTY_PAGES */ 4131c0eeaf5SJoern Engel inode->i_state |= I_SYNC; 4145547e8aaSDmitry Monakhov inode->i_state &= ~I_DIRTY_PAGES; 4151da177e4SLinus Torvalds spin_unlock(&inode_lock); 4161da177e4SLinus Torvalds 4171da177e4SLinus Torvalds ret = do_writepages(mapping, wbc); 4181da177e4SLinus Torvalds 41926821ed4SChristoph Hellwig /* 42026821ed4SChristoph Hellwig * Make sure to wait on the data before writing out the metadata. 42126821ed4SChristoph Hellwig * This is important for filesystems that modify metadata on data 42226821ed4SChristoph Hellwig * I/O completion. 42326821ed4SChristoph Hellwig */ 424a9185b41SChristoph Hellwig if (wbc->sync_mode == WB_SYNC_ALL) { 42526821ed4SChristoph Hellwig int err = filemap_fdatawait(mapping); 4261da177e4SLinus Torvalds if (ret == 0) 4271da177e4SLinus Torvalds ret = err; 4281da177e4SLinus Torvalds } 4291da177e4SLinus Torvalds 4305547e8aaSDmitry Monakhov /* 4315547e8aaSDmitry Monakhov * Some filesystems may redirty the inode during the writeback 4325547e8aaSDmitry Monakhov * due to delalloc, clear dirty metadata flags right before 4335547e8aaSDmitry Monakhov * write_inode() 4345547e8aaSDmitry Monakhov */ 4355547e8aaSDmitry Monakhov spin_lock(&inode_lock); 4365547e8aaSDmitry Monakhov dirty = inode->i_state & I_DIRTY; 4375547e8aaSDmitry Monakhov inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); 4385547e8aaSDmitry Monakhov spin_unlock(&inode_lock); 43926821ed4SChristoph Hellwig /* Don't write the inode if only I_DIRTY_PAGES was set */ 44026821ed4SChristoph Hellwig if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 441a9185b41SChristoph Hellwig int err = write_inode(inode, wbc); 4421da177e4SLinus Torvalds if (ret == 0) 4431da177e4SLinus Torvalds ret = err; 4441da177e4SLinus Torvalds } 4451da177e4SLinus Torvalds 4461da177e4SLinus Torvalds spin_lock(&inode_lock); 4471c0eeaf5SJoern Engel inode->i_state &= ~I_SYNC; 44884a89245SWu Fengguang if (!(inode->i_state & (I_FREEING | I_CLEAR))) { 449b3af9468SWu Fengguang if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) { 450ae1b7f7dSWu Fengguang /* 451b3af9468SWu Fengguang * More pages get dirtied by a fast dirtier. 452b3af9468SWu Fengguang */ 453b3af9468SWu Fengguang goto select_queue; 454b3af9468SWu Fengguang } else if (inode->i_state & I_DIRTY) { 455b3af9468SWu Fengguang /* 456b3af9468SWu Fengguang * At least XFS will redirty the inode during the 457b3af9468SWu Fengguang * writeback (delalloc) and on io completion (isize). 458ae1b7f7dSWu Fengguang */ 459ae1b7f7dSWu Fengguang redirty_tail(inode); 460ae1b7f7dSWu Fengguang } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { 4611da177e4SLinus Torvalds /* 4621da177e4SLinus Torvalds * We didn't write back all the pages. nfs_writepages() 4631da177e4SLinus Torvalds * sometimes bales out without doing anything. Redirty 46466f3b8e2SJens Axboe * the inode; Move it from b_io onto b_more_io/b_dirty. 4651b43ef91SAndrew Morton */ 4661b43ef91SAndrew Morton /* 4671b43ef91SAndrew Morton * akpm: if the caller was the kupdate function we put 46866f3b8e2SJens Axboe * this inode at the head of b_dirty so it gets first 4691b43ef91SAndrew Morton * consideration. Otherwise, move it to the tail, for 4701b43ef91SAndrew Morton * the reasons described there. I'm not really sure 4711b43ef91SAndrew Morton * how much sense this makes. Presumably I had a good 4721b43ef91SAndrew Morton * reasons for doing it this way, and I'd rather not 4731b43ef91SAndrew Morton * muck with it at present. 4741da177e4SLinus Torvalds */ 4751da177e4SLinus Torvalds if (wbc->for_kupdate) { 4761da177e4SLinus Torvalds /* 4772c136579SFengguang Wu * For the kupdate function we move the inode 47866f3b8e2SJens Axboe * to b_more_io so it will get more writeout as 4792c136579SFengguang Wu * soon as the queue becomes uncongested. 4801da177e4SLinus Torvalds */ 4811da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 482b3af9468SWu Fengguang select_queue: 4838bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 4848bc3be27SFengguang Wu /* 4858bc3be27SFengguang Wu * slice used up: queue for next turn 4868bc3be27SFengguang Wu */ 4870e0f4fc2SKen Chen requeue_io(inode); 4881da177e4SLinus Torvalds } else { 4891da177e4SLinus Torvalds /* 4908bc3be27SFengguang Wu * somehow blocked: retry later 4918bc3be27SFengguang Wu */ 4928bc3be27SFengguang Wu redirty_tail(inode); 4938bc3be27SFengguang Wu } 4948bc3be27SFengguang Wu } else { 4958bc3be27SFengguang Wu /* 4961da177e4SLinus Torvalds * Otherwise fully redirty the inode so that 4971da177e4SLinus Torvalds * other inodes on this superblock will get some 4981da177e4SLinus Torvalds * writeout. Otherwise heavy writing to one 4991da177e4SLinus Torvalds * file would indefinitely suspend writeout of 5001da177e4SLinus Torvalds * all the other files. 5011da177e4SLinus Torvalds */ 5021da177e4SLinus Torvalds inode->i_state |= I_DIRTY_PAGES; 5031b43ef91SAndrew Morton redirty_tail(inode); 5041da177e4SLinus Torvalds } 5051da177e4SLinus Torvalds } else if (atomic_read(&inode->i_count)) { 5061da177e4SLinus Torvalds /* 5071da177e4SLinus Torvalds * The inode is clean, inuse 5081da177e4SLinus Torvalds */ 5091da177e4SLinus Torvalds list_move(&inode->i_list, &inode_in_use); 5101da177e4SLinus Torvalds } else { 5111da177e4SLinus Torvalds /* 5121da177e4SLinus Torvalds * The inode is clean, unused 5131da177e4SLinus Torvalds */ 5141da177e4SLinus Torvalds list_move(&inode->i_list, &inode_unused); 5151da177e4SLinus Torvalds } 5161da177e4SLinus Torvalds } 5171c0eeaf5SJoern Engel inode_sync_complete(inode); 5181da177e4SLinus Torvalds return ret; 5191da177e4SLinus Torvalds } 5201da177e4SLinus Torvalds 52103ba3782SJens Axboe /* 522d19de7edSChristoph Hellwig * For background writeback the caller does not have the sb pinned 52303ba3782SJens Axboe * before calling writeback. So make sure that we do pin it, so it doesn't 52403ba3782SJens Axboe * go away while we are writing inodes from it. 52503ba3782SJens Axboe */ 526d19de7edSChristoph Hellwig static bool pin_sb_for_writeback(struct super_block *sb) 5271da177e4SLinus Torvalds { 52803ba3782SJens Axboe spin_lock(&sb_lock); 52903ba3782SJens Axboe sb->s_count++; 53003ba3782SJens Axboe if (down_read_trylock(&sb->s_umount)) { 53103ba3782SJens Axboe if (sb->s_root) { 53203ba3782SJens Axboe spin_unlock(&sb_lock); 533d19de7edSChristoph Hellwig return true; 53403ba3782SJens Axboe } 53503ba3782SJens Axboe /* 53603ba3782SJens Axboe * umounted, drop rwsem again and fall through to failure 53703ba3782SJens Axboe */ 53803ba3782SJens Axboe up_read(&sb->s_umount); 53903ba3782SJens Axboe } 54003ba3782SJens Axboe sb->s_count--; 54103ba3782SJens Axboe spin_unlock(&sb_lock); 542d19de7edSChristoph Hellwig return false; 54303ba3782SJens Axboe } 54403ba3782SJens Axboe 545f11c9c5cSEdward Shishkin /* 546f11c9c5cSEdward Shishkin * Write a portion of b_io inodes which belong to @sb. 547f11c9c5cSEdward Shishkin * If @wbc->sb != NULL, then find and write all such 548f11c9c5cSEdward Shishkin * inodes. Otherwise write only ones which go sequentially 549f11c9c5cSEdward Shishkin * in reverse order. 550f11c9c5cSEdward Shishkin * Return 1, if the caller writeback routine should be 551f11c9c5cSEdward Shishkin * interrupted. Otherwise return 0. 552f11c9c5cSEdward Shishkin */ 553f11c9c5cSEdward Shishkin static int writeback_sb_inodes(struct super_block *sb, 554f11c9c5cSEdward Shishkin struct bdi_writeback *wb, 55503ba3782SJens Axboe struct writeback_control *wbc) 55603ba3782SJens Axboe { 55703ba3782SJens Axboe while (!list_empty(&wb->b_io)) { 558f11c9c5cSEdward Shishkin long pages_skipped; 55903ba3782SJens Axboe struct inode *inode = list_entry(wb->b_io.prev, 5601da177e4SLinus Torvalds struct inode, i_list); 561f11c9c5cSEdward Shishkin if (wbc->sb && sb != inode->i_sb) { 562f11c9c5cSEdward Shishkin /* super block given and doesn't 563f11c9c5cSEdward Shishkin match, skip this inode */ 56466f3b8e2SJens Axboe redirty_tail(inode); 56566f3b8e2SJens Axboe continue; 56666f3b8e2SJens Axboe } 567f11c9c5cSEdward Shishkin if (sb != inode->i_sb) 568f11c9c5cSEdward Shishkin /* finish with this superblock */ 569f11c9c5cSEdward Shishkin return 0; 57084a89245SWu Fengguang if (inode->i_state & (I_NEW | I_WILL_FREE)) { 5717ef0d737SNick Piggin requeue_io(inode); 5727ef0d737SNick Piggin continue; 5737ef0d737SNick Piggin } 574d2caa3c5SJeff Layton /* 575d2caa3c5SJeff Layton * Was this inode dirtied after sync_sb_inodes was called? 576d2caa3c5SJeff Layton * This keeps sync from extra jobs and livelock. 577d2caa3c5SJeff Layton */ 578f11c9c5cSEdward Shishkin if (inode_dirtied_after(inode, wbc->wb_start)) 579f11c9c5cSEdward Shishkin return 1; 5801da177e4SLinus Torvalds 58184a89245SWu Fengguang BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); 5821da177e4SLinus Torvalds __iget(inode); 5831da177e4SLinus Torvalds pages_skipped = wbc->pages_skipped; 58401c03194SChristoph Hellwig writeback_single_inode(inode, wbc); 5851da177e4SLinus Torvalds if (wbc->pages_skipped != pages_skipped) { 5861da177e4SLinus Torvalds /* 5871da177e4SLinus Torvalds * writeback is not making progress due to locked 5881da177e4SLinus Torvalds * buffers. Skip this inode for now. 5891da177e4SLinus Torvalds */ 590f57b9b7bSAndrew Morton redirty_tail(inode); 5911da177e4SLinus Torvalds } 5921da177e4SLinus Torvalds spin_unlock(&inode_lock); 5931da177e4SLinus Torvalds iput(inode); 5944ffc8444SOGAWA Hirofumi cond_resched(); 5951da177e4SLinus Torvalds spin_lock(&inode_lock); 5968bc3be27SFengguang Wu if (wbc->nr_to_write <= 0) { 5978bc3be27SFengguang Wu wbc->more_io = 1; 598f11c9c5cSEdward Shishkin return 1; 5991da177e4SLinus Torvalds } 60003ba3782SJens Axboe if (!list_empty(&wb->b_more_io)) 6018bc3be27SFengguang Wu wbc->more_io = 1; 6028bc3be27SFengguang Wu } 603f11c9c5cSEdward Shishkin /* b_io is empty */ 604f11c9c5cSEdward Shishkin return 1; 605f11c9c5cSEdward Shishkin } 60638f21977SNick Piggin 607f11c9c5cSEdward Shishkin static void writeback_inodes_wb(struct bdi_writeback *wb, 608f11c9c5cSEdward Shishkin struct writeback_control *wbc) 609f11c9c5cSEdward Shishkin { 610f11c9c5cSEdward Shishkin int ret = 0; 6119ecc2738SJens Axboe 612f11c9c5cSEdward Shishkin wbc->wb_start = jiffies; /* livelock avoidance */ 613f11c9c5cSEdward Shishkin spin_lock(&inode_lock); 614f11c9c5cSEdward Shishkin if (!wbc->for_kupdate || list_empty(&wb->b_io)) 615f11c9c5cSEdward Shishkin queue_io(wb, wbc->older_than_this); 616f11c9c5cSEdward Shishkin 617f11c9c5cSEdward Shishkin while (!list_empty(&wb->b_io)) { 618f11c9c5cSEdward Shishkin struct inode *inode = list_entry(wb->b_io.prev, 619f11c9c5cSEdward Shishkin struct inode, i_list); 620f11c9c5cSEdward Shishkin struct super_block *sb = inode->i_sb; 621f11c9c5cSEdward Shishkin 622d19de7edSChristoph Hellwig if (wbc->sb) { 623d19de7edSChristoph Hellwig /* 624d19de7edSChristoph Hellwig * We are requested to write out inodes for a specific 625d19de7edSChristoph Hellwig * superblock. This means we already have s_umount 626d19de7edSChristoph Hellwig * taken by the caller which also waits for us to 627d19de7edSChristoph Hellwig * complete the writeout. 628d19de7edSChristoph Hellwig */ 629d19de7edSChristoph Hellwig if (sb != wbc->sb) { 630f11c9c5cSEdward Shishkin redirty_tail(inode); 631f11c9c5cSEdward Shishkin continue; 632f11c9c5cSEdward Shishkin } 633f11c9c5cSEdward Shishkin 634d19de7edSChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 635d19de7edSChristoph Hellwig 636f11c9c5cSEdward Shishkin ret = writeback_sb_inodes(sb, wb, wbc); 637d19de7edSChristoph Hellwig } else { 638d19de7edSChristoph Hellwig if (!pin_sb_for_writeback(sb)) 639d19de7edSChristoph Hellwig continue; 640d19de7edSChristoph Hellwig ret = writeback_sb_inodes(sb, wb, wbc); 641d19de7edSChristoph Hellwig drop_super(sb); 642d19de7edSChristoph Hellwig } 643f11c9c5cSEdward Shishkin 644f11c9c5cSEdward Shishkin if (ret) 645f11c9c5cSEdward Shishkin break; 646f11c9c5cSEdward Shishkin } 64766f3b8e2SJens Axboe spin_unlock(&inode_lock); 64866f3b8e2SJens Axboe /* Leave any unwritten inodes on b_io */ 64966f3b8e2SJens Axboe } 65066f3b8e2SJens Axboe 65103ba3782SJens Axboe void writeback_inodes_wbc(struct writeback_control *wbc) 65203ba3782SJens Axboe { 65303ba3782SJens Axboe struct backing_dev_info *bdi = wbc->bdi; 65403ba3782SJens Axboe 65503ba3782SJens Axboe writeback_inodes_wb(&bdi->wb, wbc); 65603ba3782SJens Axboe } 65703ba3782SJens Axboe 65803ba3782SJens Axboe /* 65903ba3782SJens Axboe * The maximum number of pages to writeout in a single bdi flush/kupdate 66003ba3782SJens Axboe * operation. We do this so we don't hold I_SYNC against an inode for 66103ba3782SJens Axboe * enormous amounts of time, which would block a userspace task which has 66203ba3782SJens Axboe * been forced to throttle against that inode. Also, the code reevaluates 66303ba3782SJens Axboe * the dirty each time it has written this many pages. 66403ba3782SJens Axboe */ 66503ba3782SJens Axboe #define MAX_WRITEBACK_PAGES 1024 66603ba3782SJens Axboe 66703ba3782SJens Axboe static inline bool over_bground_thresh(void) 66803ba3782SJens Axboe { 66903ba3782SJens Axboe unsigned long background_thresh, dirty_thresh; 67003ba3782SJens Axboe 67103ba3782SJens Axboe get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); 67203ba3782SJens Axboe 67303ba3782SJens Axboe return (global_page_state(NR_FILE_DIRTY) + 67403ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) >= background_thresh); 67503ba3782SJens Axboe } 67603ba3782SJens Axboe 67703ba3782SJens Axboe /* 67803ba3782SJens Axboe * Explicit flushing or periodic writeback of "old" data. 67903ba3782SJens Axboe * 68003ba3782SJens Axboe * Define "old": the first time one of an inode's pages is dirtied, we mark the 68103ba3782SJens Axboe * dirtying-time in the inode's address_space. So this periodic writeback code 68203ba3782SJens Axboe * just walks the superblock inode list, writing back any inodes which are 68303ba3782SJens Axboe * older than a specific point in time. 68403ba3782SJens Axboe * 68503ba3782SJens Axboe * Try to run once per dirty_writeback_interval. But if a writeback event 68603ba3782SJens Axboe * takes longer than a dirty_writeback_interval interval, then leave a 68703ba3782SJens Axboe * one-second gap. 68803ba3782SJens Axboe * 68903ba3782SJens Axboe * older_than_this takes precedence over nr_to_write. So we'll only write back 69003ba3782SJens Axboe * all dirty pages if they are all attached to "old" mappings. 69103ba3782SJens Axboe */ 692c4a77a6cSJens Axboe static long wb_writeback(struct bdi_writeback *wb, 693c4a77a6cSJens Axboe struct wb_writeback_args *args) 69403ba3782SJens Axboe { 69503ba3782SJens Axboe struct writeback_control wbc = { 69603ba3782SJens Axboe .bdi = wb->bdi, 697c4a77a6cSJens Axboe .sb = args->sb, 698c4a77a6cSJens Axboe .sync_mode = args->sync_mode, 69903ba3782SJens Axboe .older_than_this = NULL, 700c4a77a6cSJens Axboe .for_kupdate = args->for_kupdate, 701b17621feSWu Fengguang .for_background = args->for_background, 702c4a77a6cSJens Axboe .range_cyclic = args->range_cyclic, 70303ba3782SJens Axboe }; 70403ba3782SJens Axboe unsigned long oldest_jif; 70503ba3782SJens Axboe long wrote = 0; 706a5989bdcSJan Kara struct inode *inode; 70703ba3782SJens Axboe 70803ba3782SJens Axboe if (wbc.for_kupdate) { 70903ba3782SJens Axboe wbc.older_than_this = &oldest_jif; 71003ba3782SJens Axboe oldest_jif = jiffies - 71103ba3782SJens Axboe msecs_to_jiffies(dirty_expire_interval * 10); 71203ba3782SJens Axboe } 713c4a77a6cSJens Axboe if (!wbc.range_cyclic) { 714c4a77a6cSJens Axboe wbc.range_start = 0; 715c4a77a6cSJens Axboe wbc.range_end = LLONG_MAX; 716c4a77a6cSJens Axboe } 71703ba3782SJens Axboe 71803ba3782SJens Axboe for (;;) { 71903ba3782SJens Axboe /* 720d3ddec76SWu Fengguang * Stop writeback when nr_pages has been consumed 72103ba3782SJens Axboe */ 722d3ddec76SWu Fengguang if (args->nr_pages <= 0) 72303ba3782SJens Axboe break; 72403ba3782SJens Axboe 72503ba3782SJens Axboe /* 726d3ddec76SWu Fengguang * For background writeout, stop when we are below the 727d3ddec76SWu Fengguang * background dirty threshold 72803ba3782SJens Axboe */ 729d3ddec76SWu Fengguang if (args->for_background && !over_bground_thresh()) 73003ba3782SJens Axboe break; 73103ba3782SJens Axboe 73203ba3782SJens Axboe wbc.more_io = 0; 73303ba3782SJens Axboe wbc.nr_to_write = MAX_WRITEBACK_PAGES; 73403ba3782SJens Axboe wbc.pages_skipped = 0; 73503ba3782SJens Axboe writeback_inodes_wb(wb, &wbc); 736c4a77a6cSJens Axboe args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; 73703ba3782SJens Axboe wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; 73803ba3782SJens Axboe 73903ba3782SJens Axboe /* 74071fd05a8SJens Axboe * If we consumed everything, see if we have more 74103ba3782SJens Axboe */ 74271fd05a8SJens Axboe if (wbc.nr_to_write <= 0) 74371fd05a8SJens Axboe continue; 74471fd05a8SJens Axboe /* 74571fd05a8SJens Axboe * Didn't write everything and we don't have more IO, bail 74671fd05a8SJens Axboe */ 74771fd05a8SJens Axboe if (!wbc.more_io) 74871fd05a8SJens Axboe break; 74971fd05a8SJens Axboe /* 75071fd05a8SJens Axboe * Did we write something? Try for more 75171fd05a8SJens Axboe */ 752a5989bdcSJan Kara if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) 75303ba3782SJens Axboe continue; 754a5989bdcSJan Kara /* 755a5989bdcSJan Kara * Nothing written. Wait for some inode to 756a5989bdcSJan Kara * become available for writeback. Otherwise 757a5989bdcSJan Kara * we'll just busyloop. 758a5989bdcSJan Kara */ 759a5989bdcSJan Kara spin_lock(&inode_lock); 760a5989bdcSJan Kara if (!list_empty(&wb->b_more_io)) { 76171fd05a8SJens Axboe inode = list_entry(wb->b_more_io.prev, 762a5989bdcSJan Kara struct inode, i_list); 763a5989bdcSJan Kara inode_wait_for_writeback(inode); 764a5989bdcSJan Kara } 765a5989bdcSJan Kara spin_unlock(&inode_lock); 76603ba3782SJens Axboe } 76703ba3782SJens Axboe 76803ba3782SJens Axboe return wrote; 76903ba3782SJens Axboe } 77003ba3782SJens Axboe 77103ba3782SJens Axboe /* 77203ba3782SJens Axboe * Return the next bdi_work struct that hasn't been processed by this 7738010c3b6SJens Axboe * wb thread yet. ->seen is initially set for each thread that exists 7748010c3b6SJens Axboe * for this device, when a thread first notices a piece of work it 7758010c3b6SJens Axboe * clears its bit. Depending on writeback type, the thread will notify 7768010c3b6SJens Axboe * completion on either receiving the work (WB_SYNC_NONE) or after 7778010c3b6SJens Axboe * it is done (WB_SYNC_ALL). 77803ba3782SJens Axboe */ 77903ba3782SJens Axboe static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, 78003ba3782SJens Axboe struct bdi_writeback *wb) 78103ba3782SJens Axboe { 78203ba3782SJens Axboe struct bdi_work *work, *ret = NULL; 78303ba3782SJens Axboe 78403ba3782SJens Axboe rcu_read_lock(); 78503ba3782SJens Axboe 78603ba3782SJens Axboe list_for_each_entry_rcu(work, &bdi->work_list, list) { 78777fad5e6SNick Piggin if (!test_bit(wb->nr, &work->seen)) 78803ba3782SJens Axboe continue; 78977fad5e6SNick Piggin clear_bit(wb->nr, &work->seen); 79003ba3782SJens Axboe 79103ba3782SJens Axboe ret = work; 79203ba3782SJens Axboe break; 79303ba3782SJens Axboe } 79403ba3782SJens Axboe 79503ba3782SJens Axboe rcu_read_unlock(); 79603ba3782SJens Axboe return ret; 79703ba3782SJens Axboe } 79803ba3782SJens Axboe 79903ba3782SJens Axboe static long wb_check_old_data_flush(struct bdi_writeback *wb) 80003ba3782SJens Axboe { 80103ba3782SJens Axboe unsigned long expired; 80203ba3782SJens Axboe long nr_pages; 80303ba3782SJens Axboe 80469b62d01SJens Axboe /* 80569b62d01SJens Axboe * When set to zero, disable periodic writeback 80669b62d01SJens Axboe */ 80769b62d01SJens Axboe if (!dirty_writeback_interval) 80869b62d01SJens Axboe return 0; 80969b62d01SJens Axboe 81003ba3782SJens Axboe expired = wb->last_old_flush + 81103ba3782SJens Axboe msecs_to_jiffies(dirty_writeback_interval * 10); 81203ba3782SJens Axboe if (time_before(jiffies, expired)) 81303ba3782SJens Axboe return 0; 81403ba3782SJens Axboe 81503ba3782SJens Axboe wb->last_old_flush = jiffies; 81603ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 81703ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS) + 81803ba3782SJens Axboe (inodes_stat.nr_inodes - inodes_stat.nr_unused); 81903ba3782SJens Axboe 820c4a77a6cSJens Axboe if (nr_pages) { 821c4a77a6cSJens Axboe struct wb_writeback_args args = { 822c4a77a6cSJens Axboe .nr_pages = nr_pages, 823c4a77a6cSJens Axboe .sync_mode = WB_SYNC_NONE, 824c4a77a6cSJens Axboe .for_kupdate = 1, 825c4a77a6cSJens Axboe .range_cyclic = 1, 826c4a77a6cSJens Axboe }; 827c4a77a6cSJens Axboe 828c4a77a6cSJens Axboe return wb_writeback(wb, &args); 829c4a77a6cSJens Axboe } 83003ba3782SJens Axboe 83103ba3782SJens Axboe return 0; 83203ba3782SJens Axboe } 83303ba3782SJens Axboe 83403ba3782SJens Axboe /* 83503ba3782SJens Axboe * Retrieve work items and do the writeback they describe 83603ba3782SJens Axboe */ 83703ba3782SJens Axboe long wb_do_writeback(struct bdi_writeback *wb, int force_wait) 83803ba3782SJens Axboe { 83903ba3782SJens Axboe struct backing_dev_info *bdi = wb->bdi; 84003ba3782SJens Axboe struct bdi_work *work; 841c4a77a6cSJens Axboe long wrote = 0; 84203ba3782SJens Axboe 84303ba3782SJens Axboe while ((work = get_next_work_item(bdi, wb)) != NULL) { 844c4a77a6cSJens Axboe struct wb_writeback_args args = work->args; 84503ba3782SJens Axboe 84603ba3782SJens Axboe /* 84703ba3782SJens Axboe * Override sync mode, in case we must wait for completion 84803ba3782SJens Axboe */ 84903ba3782SJens Axboe if (force_wait) 850c4a77a6cSJens Axboe work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; 85103ba3782SJens Axboe 85203ba3782SJens Axboe /* 85303ba3782SJens Axboe * If this isn't a data integrity operation, just notify 85403ba3782SJens Axboe * that we have seen this work and we are now starting it. 85503ba3782SJens Axboe */ 8567f0e7bedSChristoph Hellwig if (!test_bit(WS_ONSTACK, &work->state)) 85703ba3782SJens Axboe wb_clear_pending(wb, work); 85803ba3782SJens Axboe 859c4a77a6cSJens Axboe wrote += wb_writeback(wb, &args); 86003ba3782SJens Axboe 86103ba3782SJens Axboe /* 86203ba3782SJens Axboe * This is a data integrity writeback, so only do the 86303ba3782SJens Axboe * notification when we have completed the work. 86403ba3782SJens Axboe */ 8657f0e7bedSChristoph Hellwig if (test_bit(WS_ONSTACK, &work->state)) 86603ba3782SJens Axboe wb_clear_pending(wb, work); 86703ba3782SJens Axboe } 86803ba3782SJens Axboe 86903ba3782SJens Axboe /* 87003ba3782SJens Axboe * Check for periodic writeback, kupdated() style 87103ba3782SJens Axboe */ 87203ba3782SJens Axboe wrote += wb_check_old_data_flush(wb); 87303ba3782SJens Axboe 87403ba3782SJens Axboe return wrote; 87503ba3782SJens Axboe } 87603ba3782SJens Axboe 87703ba3782SJens Axboe /* 87803ba3782SJens Axboe * Handle writeback of dirty data for the device backed by this bdi. Also 87903ba3782SJens Axboe * wakes up periodically and does kupdated style flushing. 88003ba3782SJens Axboe */ 88103ba3782SJens Axboe int bdi_writeback_task(struct bdi_writeback *wb) 88203ba3782SJens Axboe { 88303ba3782SJens Axboe unsigned long last_active = jiffies; 88403ba3782SJens Axboe unsigned long wait_jiffies = -1UL; 88503ba3782SJens Axboe long pages_written; 88603ba3782SJens Axboe 88703ba3782SJens Axboe while (!kthread_should_stop()) { 88803ba3782SJens Axboe pages_written = wb_do_writeback(wb, 0); 88903ba3782SJens Axboe 89003ba3782SJens Axboe if (pages_written) 89103ba3782SJens Axboe last_active = jiffies; 89203ba3782SJens Axboe else if (wait_jiffies != -1UL) { 89303ba3782SJens Axboe unsigned long max_idle; 89403ba3782SJens Axboe 89503ba3782SJens Axboe /* 89603ba3782SJens Axboe * Longest period of inactivity that we tolerate. If we 89703ba3782SJens Axboe * see dirty data again later, the task will get 89803ba3782SJens Axboe * recreated automatically. 89903ba3782SJens Axboe */ 90003ba3782SJens Axboe max_idle = max(5UL * 60 * HZ, wait_jiffies); 90103ba3782SJens Axboe if (time_after(jiffies, max_idle + last_active)) 90203ba3782SJens Axboe break; 90303ba3782SJens Axboe } 90403ba3782SJens Axboe 90569b62d01SJens Axboe if (dirty_writeback_interval) { 90603ba3782SJens Axboe wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); 90749db0414SJens Axboe schedule_timeout_interruptible(wait_jiffies); 908f9eadbbdSJens Axboe } else { 909f9eadbbdSJens Axboe set_current_state(TASK_INTERRUPTIBLE); 910f9eadbbdSJens Axboe if (list_empty_careful(&wb->bdi->work_list) && 911f9eadbbdSJens Axboe !kthread_should_stop()) 91269b62d01SJens Axboe schedule(); 913f9eadbbdSJens Axboe __set_current_state(TASK_RUNNING); 914f9eadbbdSJens Axboe } 91569b62d01SJens Axboe 91603ba3782SJens Axboe try_to_freeze(); 91703ba3782SJens Axboe } 91803ba3782SJens Axboe 91903ba3782SJens Axboe return 0; 92003ba3782SJens Axboe } 92103ba3782SJens Axboe 92203ba3782SJens Axboe /* 923b6e51316SJens Axboe * Schedule writeback for all backing devices. This does WB_SYNC_NONE 9243c4d7165SChristoph Hellwig * writeback, for integrity writeback see bdi_queue_work_onstack(). 92503ba3782SJens Axboe */ 926b6e51316SJens Axboe static void bdi_writeback_all(struct super_block *sb, long nr_pages) 92703ba3782SJens Axboe { 928b6e51316SJens Axboe struct wb_writeback_args args = { 929b6e51316SJens Axboe .sb = sb, 930b6e51316SJens Axboe .nr_pages = nr_pages, 931b6e51316SJens Axboe .sync_mode = WB_SYNC_NONE, 932b6e51316SJens Axboe }; 93303ba3782SJens Axboe struct backing_dev_info *bdi; 93403ba3782SJens Axboe 935cfc4ba53SJens Axboe rcu_read_lock(); 93603ba3782SJens Axboe 937cfc4ba53SJens Axboe list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { 93803ba3782SJens Axboe if (!bdi_has_dirty_io(bdi)) 93903ba3782SJens Axboe continue; 94003ba3782SJens Axboe 941f17625b3SJens Axboe bdi_alloc_queue_work(bdi, &args); 94203ba3782SJens Axboe } 94303ba3782SJens Axboe 944cfc4ba53SJens Axboe rcu_read_unlock(); 94503ba3782SJens Axboe } 94603ba3782SJens Axboe 94703ba3782SJens Axboe /* 94803ba3782SJens Axboe * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back 94903ba3782SJens Axboe * the whole world. 95003ba3782SJens Axboe */ 95103ba3782SJens Axboe void wakeup_flusher_threads(long nr_pages) 95203ba3782SJens Axboe { 95303ba3782SJens Axboe if (nr_pages == 0) 95403ba3782SJens Axboe nr_pages = global_page_state(NR_FILE_DIRTY) + 95503ba3782SJens Axboe global_page_state(NR_UNSTABLE_NFS); 956b6e51316SJens Axboe bdi_writeback_all(NULL, nr_pages); 95703ba3782SJens Axboe } 95803ba3782SJens Axboe 95903ba3782SJens Axboe static noinline void block_dump___mark_inode_dirty(struct inode *inode) 96003ba3782SJens Axboe { 96103ba3782SJens Axboe if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { 96203ba3782SJens Axboe struct dentry *dentry; 96303ba3782SJens Axboe const char *name = "?"; 96403ba3782SJens Axboe 96503ba3782SJens Axboe dentry = d_find_alias(inode); 96603ba3782SJens Axboe if (dentry) { 96703ba3782SJens Axboe spin_lock(&dentry->d_lock); 96803ba3782SJens Axboe name = (const char *) dentry->d_name.name; 96903ba3782SJens Axboe } 97003ba3782SJens Axboe printk(KERN_DEBUG 97103ba3782SJens Axboe "%s(%d): dirtied inode %lu (%s) on %s\n", 97203ba3782SJens Axboe current->comm, task_pid_nr(current), inode->i_ino, 97303ba3782SJens Axboe name, inode->i_sb->s_id); 97403ba3782SJens Axboe if (dentry) { 97503ba3782SJens Axboe spin_unlock(&dentry->d_lock); 97603ba3782SJens Axboe dput(dentry); 97703ba3782SJens Axboe } 97803ba3782SJens Axboe } 97903ba3782SJens Axboe } 98003ba3782SJens Axboe 98103ba3782SJens Axboe /** 98203ba3782SJens Axboe * __mark_inode_dirty - internal function 98303ba3782SJens Axboe * @inode: inode to mark 98403ba3782SJens Axboe * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) 98503ba3782SJens Axboe * Mark an inode as dirty. Callers should use mark_inode_dirty or 98603ba3782SJens Axboe * mark_inode_dirty_sync. 98703ba3782SJens Axboe * 98803ba3782SJens Axboe * Put the inode on the super block's dirty list. 98903ba3782SJens Axboe * 99003ba3782SJens Axboe * CAREFUL! We mark it dirty unconditionally, but move it onto the 99103ba3782SJens Axboe * dirty list only if it is hashed or if it refers to a blockdev. 99203ba3782SJens Axboe * If it was not hashed, it will never be added to the dirty list 99303ba3782SJens Axboe * even if it is later hashed, as it will have been marked dirty already. 99403ba3782SJens Axboe * 99503ba3782SJens Axboe * In short, make sure you hash any inodes _before_ you start marking 99603ba3782SJens Axboe * them dirty. 99703ba3782SJens Axboe * 99803ba3782SJens Axboe * This function *must* be atomic for the I_DIRTY_PAGES case - 99903ba3782SJens Axboe * set_page_dirty() is called under spinlock in several places. 100003ba3782SJens Axboe * 100103ba3782SJens Axboe * Note that for blockdevs, inode->dirtied_when represents the dirtying time of 100203ba3782SJens Axboe * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of 100303ba3782SJens Axboe * the kernel-internal blockdev inode represents the dirtying time of the 100403ba3782SJens Axboe * blockdev's pages. This is why for I_DIRTY_PAGES we always use 100503ba3782SJens Axboe * page->mapping->host, so the page-dirtying time is recorded in the internal 100603ba3782SJens Axboe * blockdev inode. 100703ba3782SJens Axboe */ 100803ba3782SJens Axboe void __mark_inode_dirty(struct inode *inode, int flags) 100903ba3782SJens Axboe { 101003ba3782SJens Axboe struct super_block *sb = inode->i_sb; 101103ba3782SJens Axboe 101203ba3782SJens Axboe /* 101303ba3782SJens Axboe * Don't do this for I_DIRTY_PAGES - that doesn't actually 101403ba3782SJens Axboe * dirty the inode itself 101503ba3782SJens Axboe */ 101603ba3782SJens Axboe if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { 101703ba3782SJens Axboe if (sb->s_op->dirty_inode) 101803ba3782SJens Axboe sb->s_op->dirty_inode(inode); 101903ba3782SJens Axboe } 102003ba3782SJens Axboe 102103ba3782SJens Axboe /* 102203ba3782SJens Axboe * make sure that changes are seen by all cpus before we test i_state 102303ba3782SJens Axboe * -- mikulas 102403ba3782SJens Axboe */ 102503ba3782SJens Axboe smp_mb(); 102603ba3782SJens Axboe 102703ba3782SJens Axboe /* avoid the locking if we can */ 102803ba3782SJens Axboe if ((inode->i_state & flags) == flags) 102903ba3782SJens Axboe return; 103003ba3782SJens Axboe 103103ba3782SJens Axboe if (unlikely(block_dump)) 103203ba3782SJens Axboe block_dump___mark_inode_dirty(inode); 103303ba3782SJens Axboe 103403ba3782SJens Axboe spin_lock(&inode_lock); 103503ba3782SJens Axboe if ((inode->i_state & flags) != flags) { 103603ba3782SJens Axboe const int was_dirty = inode->i_state & I_DIRTY; 103703ba3782SJens Axboe 103803ba3782SJens Axboe inode->i_state |= flags; 103903ba3782SJens Axboe 104003ba3782SJens Axboe /* 104103ba3782SJens Axboe * If the inode is being synced, just update its dirty state. 104203ba3782SJens Axboe * The unlocker will place the inode on the appropriate 104303ba3782SJens Axboe * superblock list, based upon its state. 104403ba3782SJens Axboe */ 104503ba3782SJens Axboe if (inode->i_state & I_SYNC) 104603ba3782SJens Axboe goto out; 104703ba3782SJens Axboe 104803ba3782SJens Axboe /* 104903ba3782SJens Axboe * Only add valid (hashed) inodes to the superblock's 105003ba3782SJens Axboe * dirty list. Add blockdev inodes as well. 105103ba3782SJens Axboe */ 105203ba3782SJens Axboe if (!S_ISBLK(inode->i_mode)) { 105303ba3782SJens Axboe if (hlist_unhashed(&inode->i_hash)) 105403ba3782SJens Axboe goto out; 105503ba3782SJens Axboe } 105603ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR)) 105703ba3782SJens Axboe goto out; 105803ba3782SJens Axboe 105903ba3782SJens Axboe /* 106003ba3782SJens Axboe * If the inode was already on b_dirty/b_io/b_more_io, don't 106103ba3782SJens Axboe * reposition it (that would break b_dirty time-ordering). 106203ba3782SJens Axboe */ 106303ba3782SJens Axboe if (!was_dirty) { 106403ba3782SJens Axboe struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; 1065500b067cSJens Axboe struct backing_dev_info *bdi = wb->bdi; 1066500b067cSJens Axboe 1067500b067cSJens Axboe if (bdi_cap_writeback_dirty(bdi) && 1068500b067cSJens Axboe !test_bit(BDI_registered, &bdi->state)) { 1069500b067cSJens Axboe WARN_ON(1); 1070500b067cSJens Axboe printk(KERN_ERR "bdi-%s not registered\n", 1071500b067cSJens Axboe bdi->name); 1072500b067cSJens Axboe } 107303ba3782SJens Axboe 107403ba3782SJens Axboe inode->dirtied_when = jiffies; 107503ba3782SJens Axboe list_move(&inode->i_list, &wb->b_dirty); 107603ba3782SJens Axboe } 107703ba3782SJens Axboe } 107803ba3782SJens Axboe out: 107903ba3782SJens Axboe spin_unlock(&inode_lock); 108003ba3782SJens Axboe } 108103ba3782SJens Axboe EXPORT_SYMBOL(__mark_inode_dirty); 108203ba3782SJens Axboe 108366f3b8e2SJens Axboe /* 108466f3b8e2SJens Axboe * Write out a superblock's list of dirty inodes. A wait will be performed 108566f3b8e2SJens Axboe * upon no inodes, all inodes or the final one, depending upon sync_mode. 108666f3b8e2SJens Axboe * 108766f3b8e2SJens Axboe * If older_than_this is non-NULL, then only write out inodes which 108866f3b8e2SJens Axboe * had their first dirtying at a time earlier than *older_than_this. 108966f3b8e2SJens Axboe * 109066f3b8e2SJens Axboe * If `bdi' is non-zero then we're being asked to writeback a specific queue. 109166f3b8e2SJens Axboe * This function assumes that the blockdev superblock's inodes are backed by 109266f3b8e2SJens Axboe * a variety of queues, so all inodes are searched. For other superblocks, 109366f3b8e2SJens Axboe * assume that all inodes are backed by the same queue. 109466f3b8e2SJens Axboe * 109566f3b8e2SJens Axboe * The inodes to be written are parked on bdi->b_io. They are moved back onto 109666f3b8e2SJens Axboe * bdi->b_dirty as they are selected for writing. This way, none can be missed 109766f3b8e2SJens Axboe * on the writer throttling path, and we get decent balancing between many 109866f3b8e2SJens Axboe * throttled threads: we don't want them all piling up on inode_sync_wait. 109966f3b8e2SJens Axboe */ 1100b6e51316SJens Axboe static void wait_sb_inodes(struct super_block *sb) 110166f3b8e2SJens Axboe { 110238f21977SNick Piggin struct inode *inode, *old_inode = NULL; 110338f21977SNick Piggin 110403ba3782SJens Axboe /* 110503ba3782SJens Axboe * We need to be protected against the filesystem going from 110603ba3782SJens Axboe * r/o to r/w or vice versa. 110703ba3782SJens Axboe */ 1108b6e51316SJens Axboe WARN_ON(!rwsem_is_locked(&sb->s_umount)); 110903ba3782SJens Axboe 111066f3b8e2SJens Axboe spin_lock(&inode_lock); 111166f3b8e2SJens Axboe 111238f21977SNick Piggin /* 111338f21977SNick Piggin * Data integrity sync. Must wait for all pages under writeback, 111438f21977SNick Piggin * because there may have been pages dirtied before our sync 111538f21977SNick Piggin * call, but which had writeout started before we write it out. 111638f21977SNick Piggin * In which case, the inode may not be on the dirty list, but 111738f21977SNick Piggin * we still have to wait for that writeout. 111838f21977SNick Piggin */ 1119b6e51316SJens Axboe list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 112038f21977SNick Piggin struct address_space *mapping; 112138f21977SNick Piggin 112203ba3782SJens Axboe if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) 112338f21977SNick Piggin continue; 112438f21977SNick Piggin mapping = inode->i_mapping; 112538f21977SNick Piggin if (mapping->nrpages == 0) 112638f21977SNick Piggin continue; 112738f21977SNick Piggin __iget(inode); 1128ae8547b0SHans Reiser spin_unlock(&inode_lock); 112938f21977SNick Piggin /* 113038f21977SNick Piggin * We hold a reference to 'inode' so it couldn't have 113138f21977SNick Piggin * been removed from s_inodes list while we dropped the 113238f21977SNick Piggin * inode_lock. We cannot iput the inode now as we can 113338f21977SNick Piggin * be holding the last reference and we cannot iput it 113438f21977SNick Piggin * under inode_lock. So we keep the reference and iput 113538f21977SNick Piggin * it later. 113638f21977SNick Piggin */ 113738f21977SNick Piggin iput(old_inode); 113838f21977SNick Piggin old_inode = inode; 113938f21977SNick Piggin 114038f21977SNick Piggin filemap_fdatawait(mapping); 114138f21977SNick Piggin 114238f21977SNick Piggin cond_resched(); 114338f21977SNick Piggin 114438f21977SNick Piggin spin_lock(&inode_lock); 114538f21977SNick Piggin } 114638f21977SNick Piggin spin_unlock(&inode_lock); 114738f21977SNick Piggin iput(old_inode); 114866f3b8e2SJens Axboe } 11491da177e4SLinus Torvalds 1150d8a8559cSJens Axboe /** 1151d8a8559cSJens Axboe * writeback_inodes_sb - writeback dirty inodes from given super_block 1152d8a8559cSJens Axboe * @sb: the superblock 11531da177e4SLinus Torvalds * 1154d8a8559cSJens Axboe * Start writeback on some inodes on this super_block. No guarantees are made 1155d8a8559cSJens Axboe * on how many (if any) will be written, and this function does not wait 1156d8a8559cSJens Axboe * for IO completion of submitted IO. The number of pages submitted is 1157d8a8559cSJens Axboe * returned. 11581da177e4SLinus Torvalds */ 1159b6e51316SJens Axboe void writeback_inodes_sb(struct super_block *sb) 11601da177e4SLinus Torvalds { 11610e3c9a22SJens Axboe unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); 11620e3c9a22SJens Axboe unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); 11633c4d7165SChristoph Hellwig struct wb_writeback_args args = { 11643c4d7165SChristoph Hellwig .sb = sb, 11653c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_NONE, 11663c4d7165SChristoph Hellwig }; 11670e3c9a22SJens Axboe 1168cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1169cf37e972SChristoph Hellwig 11703c4d7165SChristoph Hellwig args.nr_pages = nr_dirty + nr_unstable + 11710e3c9a22SJens Axboe (inodes_stat.nr_inodes - inodes_stat.nr_unused); 11720e3c9a22SJens Axboe 11733c4d7165SChristoph Hellwig bdi_queue_work_onstack(&args); 11741da177e4SLinus Torvalds } 1175d8a8559cSJens Axboe EXPORT_SYMBOL(writeback_inodes_sb); 1176d8a8559cSJens Axboe 1177d8a8559cSJens Axboe /** 117817bd55d0SEric Sandeen * writeback_inodes_sb_if_idle - start writeback if none underway 117917bd55d0SEric Sandeen * @sb: the superblock 118017bd55d0SEric Sandeen * 118117bd55d0SEric Sandeen * Invoke writeback_inodes_sb if no writeback is currently underway. 118217bd55d0SEric Sandeen * Returns 1 if writeback was started, 0 if not. 118317bd55d0SEric Sandeen */ 118417bd55d0SEric Sandeen int writeback_inodes_sb_if_idle(struct super_block *sb) 118517bd55d0SEric Sandeen { 118617bd55d0SEric Sandeen if (!writeback_in_progress(sb->s_bdi)) { 1187cf37e972SChristoph Hellwig down_read(&sb->s_umount); 118817bd55d0SEric Sandeen writeback_inodes_sb(sb); 1189cf37e972SChristoph Hellwig up_read(&sb->s_umount); 119017bd55d0SEric Sandeen return 1; 119117bd55d0SEric Sandeen } else 119217bd55d0SEric Sandeen return 0; 119317bd55d0SEric Sandeen } 119417bd55d0SEric Sandeen EXPORT_SYMBOL(writeback_inodes_sb_if_idle); 119517bd55d0SEric Sandeen 119617bd55d0SEric Sandeen /** 1197d8a8559cSJens Axboe * sync_inodes_sb - sync sb inode pages 1198d8a8559cSJens Axboe * @sb: the superblock 1199d8a8559cSJens Axboe * 1200d8a8559cSJens Axboe * This function writes and waits on any dirty inode belonging to this 1201d8a8559cSJens Axboe * super_block. The number of pages synced is returned. 1202d8a8559cSJens Axboe */ 1203b6e51316SJens Axboe void sync_inodes_sb(struct super_block *sb) 1204d8a8559cSJens Axboe { 12053c4d7165SChristoph Hellwig struct wb_writeback_args args = { 12063c4d7165SChristoph Hellwig .sb = sb, 12073c4d7165SChristoph Hellwig .sync_mode = WB_SYNC_ALL, 12083c4d7165SChristoph Hellwig .nr_pages = LONG_MAX, 12093c4d7165SChristoph Hellwig .range_cyclic = 0, 12103c4d7165SChristoph Hellwig }; 12113c4d7165SChristoph Hellwig 1212cf37e972SChristoph Hellwig WARN_ON(!rwsem_is_locked(&sb->s_umount)); 1213cf37e972SChristoph Hellwig 12143c4d7165SChristoph Hellwig bdi_queue_work_onstack(&args); 1215b6e51316SJens Axboe wait_sb_inodes(sb); 1216d8a8559cSJens Axboe } 1217d8a8559cSJens Axboe EXPORT_SYMBOL(sync_inodes_sb); 12181da177e4SLinus Torvalds 12191da177e4SLinus Torvalds /** 12201da177e4SLinus Torvalds * write_inode_now - write an inode to disk 12211da177e4SLinus Torvalds * @inode: inode to write to disk 12221da177e4SLinus Torvalds * @sync: whether the write should be synchronous or not 12231da177e4SLinus Torvalds * 12247f04c26dSAndrea Arcangeli * This function commits an inode to disk immediately if it is dirty. This is 12257f04c26dSAndrea Arcangeli * primarily needed by knfsd. 12267f04c26dSAndrea Arcangeli * 12277f04c26dSAndrea Arcangeli * The caller must either have a ref on the inode or must have set I_WILL_FREE. 12281da177e4SLinus Torvalds */ 12291da177e4SLinus Torvalds int write_inode_now(struct inode *inode, int sync) 12301da177e4SLinus Torvalds { 12311da177e4SLinus Torvalds int ret; 12321da177e4SLinus Torvalds struct writeback_control wbc = { 12331da177e4SLinus Torvalds .nr_to_write = LONG_MAX, 123418914b18SMike Galbraith .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, 1235111ebb6eSOGAWA Hirofumi .range_start = 0, 1236111ebb6eSOGAWA Hirofumi .range_end = LLONG_MAX, 12371da177e4SLinus Torvalds }; 12381da177e4SLinus Torvalds 12391da177e4SLinus Torvalds if (!mapping_cap_writeback_dirty(inode->i_mapping)) 124049364ce2SAndrew Morton wbc.nr_to_write = 0; 12411da177e4SLinus Torvalds 12421da177e4SLinus Torvalds might_sleep(); 12431da177e4SLinus Torvalds spin_lock(&inode_lock); 124401c03194SChristoph Hellwig ret = writeback_single_inode(inode, &wbc); 12451da177e4SLinus Torvalds spin_unlock(&inode_lock); 12461da177e4SLinus Torvalds if (sync) 12471c0eeaf5SJoern Engel inode_sync_wait(inode); 12481da177e4SLinus Torvalds return ret; 12491da177e4SLinus Torvalds } 12501da177e4SLinus Torvalds EXPORT_SYMBOL(write_inode_now); 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds /** 12531da177e4SLinus Torvalds * sync_inode - write an inode and its pages to disk. 12541da177e4SLinus Torvalds * @inode: the inode to sync 12551da177e4SLinus Torvalds * @wbc: controls the writeback mode 12561da177e4SLinus Torvalds * 12571da177e4SLinus Torvalds * sync_inode() will write an inode and its pages to disk. It will also 12581da177e4SLinus Torvalds * correctly update the inode on its superblock's dirty inode lists and will 12591da177e4SLinus Torvalds * update inode->i_state. 12601da177e4SLinus Torvalds * 12611da177e4SLinus Torvalds * The caller must have a ref on the inode. 12621da177e4SLinus Torvalds */ 12631da177e4SLinus Torvalds int sync_inode(struct inode *inode, struct writeback_control *wbc) 12641da177e4SLinus Torvalds { 12651da177e4SLinus Torvalds int ret; 12661da177e4SLinus Torvalds 12671da177e4SLinus Torvalds spin_lock(&inode_lock); 126801c03194SChristoph Hellwig ret = writeback_single_inode(inode, wbc); 12691da177e4SLinus Torvalds spin_unlock(&inode_lock); 12701da177e4SLinus Torvalds return ret; 12711da177e4SLinus Torvalds } 12721da177e4SLinus Torvalds EXPORT_SYMBOL(sync_inode); 1273